Automated Medical Classification of Human Brain Tumors Leveraging the Xception Convolutional Neural Network

Abstract

Brain tumors are among the most critical neurological disorders, marked by abnormal cell proliferation within the brain, either benign or malignant,adversely impacting cognitive, motor, and overall patient quality of life.Accurate and prompt diagnosis is pivotal for determining effective treatment and improving survival outcomes. While Magnetic Resonance Imaging (MIRI) remains the standard diagnostic tool due to its high soft-tissue contrast, manual interpretation is labor-intensive, expertise-dependent, and subject to observer bias. Consequently, deep learning approaches, particularly Convolutional Neural Networks (CNN), have garnered considerable attention for automating brain tumor classification with superior efficiency and accuracy. This study presents a medical classification model for human brain tumors based on the Xception CNN architecture. The model was developed using a publicly available MRI dataset comprising 2,875 images categorized into glioma, meningioma, and pituitary tumor classes. Preprocessing involved resizing, normalization, and data augmentation. The model was initialized with ImageNet weights and fine-tuned for the three-class classification task with softmax activation.The proposed model achieved robust performance, recording test accuracy of 98.4% and an average F1-score of 98.5%, indicating balanced precision and recall. Confusion matrix and error analysis revealed minimal and evenly distributed misclassifications, while training dynamics showed rapid convergence with no significant overfitting.These findings demonstrate the effectiveness and clinical feasibility of the Xception CNN for automated brain tumor diagnosis. Future research should validate the approach on larger, multi-institutional datasets and integrate interpretability techniques to strengthen clinical applicability.